You ask an AI assistant for businesses in your category. Competitors appear. Your brand doesn't. Then you ask again, slightly differently, and get another list. Before rewriting your website, establish what is actually missing and whether the result points to a fixable problem.
A brand not appearing in AI answers can reflect blocked access, unclear business information, a poor match for the question, or simply the limits of the answers you checked. This troubleshooting guide helps you separate those possibilities. It doesn't assume that every absence is a penalty or that a particular change will earn a recommendation.
TLDR
For business owners and marketers, the first job is to classify the missing appearance. Can the platform identify your business when asked directly? Does it mention you without citing your website? Is the problem limited to one platform? Save a small, fixed set of relevant questions, check access and identity, then investigate the pages and sources behind the answers. My recommendation is to repair confirmed defects before buying a tool or commissioning more content. Eligibility makes appearance possible; it does not guarantee selection.
Define the missing appearance before diagnosing it
Keep three outcomes separate: a brand mention, a recommendation, and a citation linking to your website. An answer can recommend a business using a third-party source. It can also cite your educational article without recommending your services. Our guide to AI visibility explains those distinctions.
Save the exact question, complete answer, citations, platform, mode, date, and relevant market context. Note whether web search was used. Don't compare an answer generated without current web retrieval with a search-grounded answer as though both tested the same thing. The AI search guide provides the broader background.
Choose questions that reflect your actual customers and service area. Separate direct identity checks, such as asking what your company does, from non-branded discovery questions about suitable providers. Keep the wording fixed across ChatGPT search, Google AI Mode, and Perplexity, and repeat a subset where useful. Record unavailable checks as unmeasured, not as failures.
| Observed symptom | Investigate first | Do not conclude yet |
|---|---|---|
| Direct brand question gets facts wrong | Business identity and the sources cited | That more blog posts will correct the record |
| Brand appears but your site is not cited | Whether another source supports the mention | That the brand is invisible |
| Only one platform misses the brand | Its access rules and the specific mode tested | That all AI systems reject the business |
| Branded questions work but discovery does not | Fit for the buying question and available evidence | That there must be a technical block |
| One saved answer changes on recheck | Repeatability and comparable conditions | That a recent edit caused the change |
Check access when the website cannot be found
Start with a priority service or product page, not just the homepage. Ask your developer to check response codes, accidental noindex directives, robots rules, canonical signals, and whether important content is available to crawlers. An ordinary browser loading successfully doesn't prove that verified crawlers get the same response.
For Google, inspect both the indexed URL information and a live test in Search Console. The indexed report shows Google's selected canonical. The live test can help verify a repair, but it cannot establish every indexing condition or guarantee indexing (source). Record the affected URL and evidence rather than relying on a generic crawlability score.
Google says a page needs to be indexed and eligible for a Search snippet to appear in its generative Search features. The site must also be included through the Search generative AI control. Meeting the requirements does not guarantee that the content will be served (source).
Check Settings > Search generative AI in Search Console, including any inherited property setting. Google's documentation says the control reached websites worldwide on August 31, 2026. An exclusion affects specified generative features without changing ranking or inclusion in other parts of Search. That makes it a relevant check when ordinary Google visibility remains but AI-feature visibility is missing (source). Preserve a deliberate owner exclusion unless the owner approves a change.
Investigate platform specific blocks separately
OpenAI distinguishes OAI-SearchBot, used for ChatGPT search, from GPTBot, associated with model training. Those controls are independent. Its documentation says sites opting out of OAI-SearchBot won't be shown in ChatGPT search answers, although navigational links can still appear (source). A training opt-out alone is therefore not evidence of a search opt-out.
Perplexity recommends allowing PerplexityBot in robots.txt and permitting requests from its published IP ranges. Its firewall guidance combines user-agent and IP checks (source). Have the developer inspect verified request logs and security challenges. Don't disable the firewall globally or trust a user-agent string alone.
If a block is confirmed, document the intended policy, change the narrow rule responsible, and verify access afterward. Keep the technical repair separate from the later visibility recheck. Successful access establishes that the barrier was removed, not that an answer must now cite the page.
Correct identity when the assistant finds the wrong business
If the answer confuses your company with a similarly named business, follow its citations. Compare the cited records with your website's business name, domain, location, services, and public profiles. Identify which statements are wrong and where those statements originate.
Make the website's About, contact, and relevant offer pages explicit about the business. Reconcile outdated names, moved locations, discontinued products, and inconsistent service areas in profiles you control. Request factual corrections from third-party publishers when appropriate. Keep a record of the source corrected and the date; a correction request is not proof that the publisher has updated it.
Consider an illustrative situation: an assistant associates a consulting firm with an old city because a directory still lists its former address. The next task is to verify and correct that record, not publish a generic article about the new city. The example describes a diagnostic approach, not a GrowthGary client result.
Check relevance when the brand is known but not recommended
A correct answer to a direct brand question only establishes that the system can identify the business in that check. It doesn't prove the business belongs in every recommendation. Review the omitted discovery question for constraints such as geography, customer size, budget, availability, or a required specialization.
Inspect the page that should demonstrate the fit. Does it actually explain the requested service, who qualifies, and any important limitations? If a question asks for a provider offering evening appointments, vague claims about flexibility don't establish that fact. Add verified information to the appropriate existing page when the service genuinely qualifies.
Next examine cited sources and named competitors separately. A directory or trade publication may support an answer without being your competitor. Look for a concrete information gap or inaccurate listing, not a supposed secret ranking formula. Treat the comparison as a hypothesis to investigate; the answer doesn't reveal exactly why the platform selected each business.
Use real evidence where it helps a buyer decide: accurate specifications, credentials, documented service details, or permissioned project examples. Don't invent reviews, testing, results, or experience to make the page appear more authoritative.
Avoid fixes that do not match the evidence
Missing visibility alone doesn't establish that you need more schema, hundreds of new pages, or an AI text file. Google's guide says Search doesn't use llms.txt and doesn't require content to be broken into tiny pieces. It also states that there is no ideal page length (source). That guidance concerns Google Search; don't turn it into a claim about every other service.
My take is simple: ask what observation a proposed fix addresses. A confirmed crawler block justifies an access repair. A missing service explanation justifies a page update. A single disappointing answer does not justify rebuilding the website.
Verify the repair and recheck the original questions
Keep a short issue log with the evidence, affected page or profile, next action, owner, and verification method. Separate confirmed defects from unproven explanations. Fix the clearest barrier first so you can tell what was changed.
After the technical or factual repair is verified, repeat the saved questions under comparable conditions. Preserve the new answers and compare mentions, recommendations, citations, and factual accuracy separately. If results differ, describe the observed change without claiming that one edit caused it. If nothing changes, return to the next supported hypothesis rather than repeatedly changing the prompts until the brand appears.
For a broader plan after diagnosis, use our guide on how to improve AI visibility. If you need help organizing the investigation, the GrowthGary AI SEO Tools Suite outlines focused report paths. Check the current availability and scope against the problem you've identified; a report should provide evidence and next steps, not a promised citation.
You may finish with no confirmed defect. That is a legitimate outcome. Keep a dated baseline, monitor questions that matter to customers, and improve the weakest verified information. The useful result is a clearer diagnosis, even when the honest answer is that the reason for omission remains unknown.